{
  "id": 3388283,
  "title": "Which Skill Is Quietly Burning Your Tokens? Find Out From transcript.jsonl",
  "url": "https://urgent.news/2026/08/26/which-skill-is-quietly-burning-your-tokens-find-out-from-transcript",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-08-26T00:42:35.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/bokuwalily/which-skill-is-quietly-burning-your-tokens-find-out-from-transcriptjsonl-n65"
  },
  "original_language": "en",
  "account": "The script usage-breakdown.sh is a 106-line shell script that helps identify which Skills are consuming the most tokens in Claude Code bills. When the monthly Claude Code bill goes up unexpectedly, as in a 20% increase, it's difficult to pinpoint which Skill caused the rise using the tooling provided. This script simplifies the process by parsing transcript.jsonl files and tallying call counts per Skill, Agent, and MCP server.\n\nClaude Code streams every operation during a session into .jsonl files under ~/.claude/projects/. The files are JSONL, with one event per line and one file per session. The structure of a single record includes a message with a role and content, where the content field contains tool_use blocks. The name field in these blocks indicates the tool invoked, such as Skill, Agent, or MCP calls.\n\nThe script runs through a Counter to analyze the data, revealing which Skills are responsible for the increased costs. While the /usage command provides a per-model cost total, it lacks the breakdown of the cost by session, Skill, and call frequency. usage-breakdown.sh focuses on tallying call counts to outline what's heavy, as the tokens consumed by a Skill called 100 times differ significantly from those called once.\n\nThe script allows users to narrow the analysis to the last N days using an mtime window. By default, it analyzes the last 7 days, but the user can adjust this to 30 or 14 days. With the default setting, the script outputs a one-line summary suitable for a status line. Additionally, the script separates the counters for different types of calls, such as Skill, Agent, and MCP servers, enabling a clear understanding of where the token consumption is coming from.",
  "summary": "Your monthly Claude Code bill went up 20%. You know that much. What you don't know is which Skill did it — and nothing in the tooling will tell you. Run /usage in Claude Code and you get claude-sonnet-4-6: ¥3,240 — a per-model total and nothing else . \"More expensive than last week\" is visible. \"Which Skill caused it\" is not. usage-breakdown.sh closes that gap. It's a 106-line shell script that…",
  "key_points": [
    "\"scriptusage-breakdown.sh\" identifies top token-consuming Skills",
    "Parses Claude Code's .jsonl files to tally call counts",
    "Default analysis covers last 7 days, adjustable to 30 or 14"
  ],
  "editors_take": "This script helps users identify and manage token consumption by Skills, Agents, and MCP servers, providing a more detailed breakdown than existing tooling, and enabling more targeted cost control.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}